Demangling in backtraces

Demangling in backtraces

[Eli Zaretskii]

I have built a C program linked against libgccjit.  When it crashes,
the backtrace shows mangled names, like this:

 #0  0x70f5ac3e in libgccjit-0!_Z17gimple_build_callP9tree_nodejz ()
    from D:\usr\bin\libgccjit-0.dll
 #1  0x7190fa7b in libgccjit-0!_ZN19evrp_range_analyzer5leaveEP15basic_block_def
 () from D:\usr\bin\libgccjit-0.dll
 #2  0x71910eef in libgccjit-0!_Z36stmt_uses_0_or_null_in_undefined_wayP6gimple
     () from D:\usr\bin\libgccjit-0.dll
 #3  0x710fba2c in libgccjit-0!_Z16execute_one_passP8opt_pass ()
    from D:\usr\bin\libgccjit-0.dll

(This is because libgccjit is written in C++, and because the DLL is
stripped.)

How can I cause GDB to demangle the function names when showing the
backtrace in this case?  I tried setting language to c++, but that
didn't help.

TIA

Re: Demangling in backtraces

[Eli Zaretskii]

Ping!  Does the silence mean we don't have such a feature in GDB?

> Date: Sat, 06 Mar 2021 20:56:13 +0200
> From: Eli Zaretskii via Gdb <gdb@sourceware.org>
> 
> I have built a C program linked against libgccjit.  When it crashes,
> the backtrace shows mangled names, like this:
> 
>  #0  0x70f5ac3e in libgccjit-0!_Z17gimple_build_callP9tree_nodejz ()
>     from D:\usr\bin\libgccjit-0.dll
>  #1  0x7190fa7b in libgccjit-0!_ZN19evrp_range_analyzer5leaveEP15basic_block_def
>  () from D:\usr\bin\libgccjit-0.dll
>  #2  0x71910eef in libgccjit-0!_Z36stmt_uses_0_or_null_in_undefined_wayP6gimple
>      () from D:\usr\bin\libgccjit-0.dll
>  #3  0x710fba2c in libgccjit-0!_Z16execute_one_passP8opt_pass ()
>     from D:\usr\bin\libgccjit-0.dll
> 
> (This is because libgccjit is written in C++, and because the DLL is
> stripped.)
> 
> How can I cause GDB to demangle the function names when showing the
> backtrace in this case?  I tried setting language to c++, but that
> didn't help.
> 
> TIA
> 

Re: Demangling in backtraces

[Simon Marchi]

On 2021-03-13 3:55 a.m., Eli Zaretskii via Gdb wrote:
> Ping!  Does the silence mean we don't have such a feature in GDB?
> 
>> Date: Sat, 06 Mar 2021 20:56:13 +0200
>> From: Eli Zaretskii via Gdb <gdb@sourceware.org>
>>
>> I have built a C program linked against libgccjit.  When it crashes,
>> the backtrace shows mangled names, like this:
>>
>>  #0  0x70f5ac3e in libgccjit-0!_Z17gimple_build_callP9tree_nodejz ()
>>     from D:\usr\bin\libgccjit-0.dll
>>  #1  0x7190fa7b in libgccjit-0!_ZN19evrp_range_analyzer5leaveEP15basic_block_def
>>  () from D:\usr\bin\libgccjit-0.dll
>>  #2  0x71910eef in libgccjit-0!_Z36stmt_uses_0_or_null_in_undefined_wayP6gimple
>>      () from D:\usr\bin\libgccjit-0.dll
>>  #3  0x710fba2c in libgccjit-0!_Z16execute_one_passP8opt_pass ()
>>     from D:\usr\bin\libgccjit-0.dll
>>
>> (This is because libgccjit is written in C++, and because the DLL is
>> stripped.)
>>
>> How can I cause GDB to demangle the function names when showing the
>> backtrace in this case?  I tried setting language to c++, but that
>> didn't help.
>>
>> TIA
>>

I don't really know.  I've tried making a toy program and shared library
on Linux in C++, without debug symbols, and it "just works":

    #0  0x00007ffff7fc00ed in foo(int) () from ./libfoo.so
    #1  0x0000555555555147 in main ()

So maybe it's supposed to "just work" on Windows as well and it doesn't?
I don't really have time to dig into it further.

Simon

Re: Demangling in backtraces

[Eli Zaretskii]

> From: Simon Marchi <simon.marchi@polymtl.ca>
> Date: Sat, 13 Mar 2021 21:46:32 -0500
> 
> >> I have built a C program linked against libgccjit.  When it crashes,
> >> the backtrace shows mangled names, like this:
> >>
> >>  #0  0x70f5ac3e in libgccjit-0!_Z17gimple_build_callP9tree_nodejz ()
> >>     from D:\usr\bin\libgccjit-0.dll
> >>  #1  0x7190fa7b in libgccjit-0!_ZN19evrp_range_analyzer5leaveEP15basic_block_def
> >>  () from D:\usr\bin\libgccjit-0.dll
> >>  #2  0x71910eef in libgccjit-0!_Z36stmt_uses_0_or_null_in_undefined_wayP6gimple
> >>      () from D:\usr\bin\libgccjit-0.dll
> >>  #3  0x710fba2c in libgccjit-0!_Z16execute_one_passP8opt_pass ()
> >>     from D:\usr\bin\libgccjit-0.dll
> >>
> >> (This is because libgccjit is written in C++, and because the DLL is
> >> stripped.)
> >>
> >> How can I cause GDB to demangle the function names when showing the
> >> backtrace in this case?  I tried setting language to c++, but that
> >> didn't help.
> >>
> >> TIA
> >>
> 
> I don't really know.  I've tried making a toy program and shared library
> on Linux in C++, without debug symbols, and it "just works":
> 
>     #0  0x00007ffff7fc00ed in foo(int) () from ./libfoo.so
>     #1  0x0000555555555147 in main ()
> 
> So maybe it's supposed to "just work" on Windows as well and it doesn't?
> I don't really have time to dig into it further.

Thanks.  So does this mean we don't have any kind of "manual
demangler" in GDB that could be turned on in cases like this, when
"just works" doesn't?  Hannes, any ideas?

Re: Demangling in backtraces

[Andreas Schwab]

On Mär 06 2021, Eli Zaretskii via Gdb wrote:

> I have built a C program linked against libgccjit.  When it crashes,
> the backtrace shows mangled names, like this:
>
>  #0  0x70f5ac3e in libgccjit-0!_Z17gimple_build_callP9tree_nodejz ()

Is this prefix part of the symbol name?  A valid mangled name must start
with _Z.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

Re: Demangling in backtraces

[Andreas Schwab]

On Mär 14 2021, Eli Zaretskii via Gdb wrote:

> Thanks.  So does this mean we don't have any kind of "manual
> demangler" in GDB that could be turned on in cases like this, when
> "just works" doesn't?  Hannes, any ideas?

(gdb) help demangle 
Demangle a mangled name.
Usage: demangle [-l LANGUAGE] [--] NAME
If LANGUAGE is not specified, NAME is demangled in the current language.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

Re: Demangling in backtraces

[Eli Zaretskii]

> From: Andreas Schwab <schwab@linux-m68k.org>
> Cc: Eli Zaretskii <eliz@gnu.org>
> Date: Sun, 14 Mar 2021 08:28:36 +0100
> 
> >  #0  0x70f5ac3e in libgccjit-0!_Z17gimple_build_callP9tree_nodejz ()
> 
> Is this prefix part of the symbol name?  A valid mangled name must start
> with _Z.

No, that's the name of the shared library.  This is a known
peculiarity on Windows, see "Non-debug DLL Symbols" in the GDB manual.

Are you saying that if those prefixes weren't there, GDB would have
demangled the names automatically?

Re: Demangling in backtraces

[Eli Zaretskii]

> From: Andreas Schwab <schwab@linux-m68k.org>
> Cc: Simon Marchi <simon.marchi@polymtl.ca>,  Hannes Domani
>  <ssbssa@yahoo.de>,  Eli Zaretskii <eliz@gnu.org>
> Date: Sun, 14 Mar 2021 08:29:59 +0100
> 
> On Mär 14 2021, Eli Zaretskii via Gdb wrote:
> 
> > Thanks.  So does this mean we don't have any kind of "manual
> > demangler" in GDB that could be turned on in cases like this, when
> > "just works" doesn't?  Hannes, any ideas?
> 
> (gdb) help demangle 
> Demangle a mangled name.
> Usage: demangle [-l LANGUAGE] [--] NAME
> If LANGUAGE is not specified, NAME is demangled in the current language.

Sure, I know.  I thought there could be some "set FOO" knob that would
demangle automatically in the backtraces.

The above works, of course (one must use "-l c++", btw), but not if I
submit the "decorated" name, such as
"libgccjit-0!_Z17gimple_build_callP9tree_nodejz".

Thanks.

Re: Demangling in backtraces

[Jeffrey Walton]

On Sun, Mar 14, 2021 at 3:48 AM Andreas Schwab <schwab@linux-m68k.org> wrote:
>
> On Mär 06 2021, Eli Zaretskii via Gdb wrote:
>
> > I have built a C program linked against libgccjit.  When it crashes,
> > the backtrace shows mangled names, like this:
> >
> >  #0  0x70f5ac3e in libgccjit-0!_Z17gimple_build_callP9tree_nodejz ()
>
> Is this prefix part of the symbol name?  A valid mangled name must start
> with _Z.

It looks like the prefix 'libgccjit-0!' is causing trouble:

$ echo '_Z17gimple_build_callP9tree_nodejz' | c++filt
gimple_build_call(tree_node*, unsigned int, ...)

$ echo 'libgccjit-0!_Z17gimple_build_callP9tree_nodejz'
libgccjit-0!_Z17gimple_build_callP9tree_nodejz

Jeff

Re: Demangling in backtraces

[Nagmat Nazarov]

How to unsubscribe from gdb@sourceware.org ?

On Sun, Mar 14, 2021 at 5:04 PM Eli Zaretskii via Gdb <gdb@sourceware.org>
wrote:

> > From: Andreas Schwab <schwab@linux-m68k.org>
> > Cc: Simon Marchi <simon.marchi@polymtl.ca>,  Hannes Domani
> >  <ssbssa@yahoo.de>,  Eli Zaretskii <eliz@gnu.org>
> > Date: Sun, 14 Mar 2021 08:29:59 +0100
> >
> > On Mär 14 2021, Eli Zaretskii via Gdb wrote:
> >
> > > Thanks.  So does this mean we don't have any kind of "manual
> > > demangler" in GDB that could be turned on in cases like this, when
> > > "just works" doesn't?  Hannes, any ideas?
> >
> > (gdb) help demangle
> > Demangle a mangled name.
> > Usage: demangle [-l LANGUAGE] [--] NAME
> > If LANGUAGE is not specified, NAME is demangled in the current language.
>
> Sure, I know.  I thought there could be some "set FOO" knob that would
> demangle automatically in the backtraces.
>
> The above works, of course (one must use "-l c++", btw), but not if I
> submit the "decorated" name, such as
> "libgccjit-0!_Z17gimple_build_callP9tree_nodejz".
>
> Thanks.
>

Re: Demangling in backtraces

[Andreas Schwab]

On Mär 14 2021, Eli Zaretskii wrote:

>> From: Andreas Schwab <schwab@linux-m68k.org>
>> Cc: Eli Zaretskii <eliz@gnu.org>
>> Date: Sun, 14 Mar 2021 08:28:36 +0100
>> 
>> >  #0  0x70f5ac3e in libgccjit-0!_Z17gimple_build_callP9tree_nodejz ()
>> 
>> Is this prefix part of the symbol name?  A valid mangled name must start
>> with _Z.
>
> No, that's the name of the shared library.  This is a known
> peculiarity on Windows, see "Non-debug DLL Symbols" in the GDB manual.

So it _is_ part of the symbol name?

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

Re: Demangling in backtraces

[Eli Zaretskii]

> From: Andreas Schwab <schwab@linux-m68k.org>
> Cc: gdb@sourceware.org
> Date: Sun, 14 Mar 2021 09:26:02 +0100
> 
> On Mär 14 2021, Eli Zaretskii wrote:
> 
> >> From: Andreas Schwab <schwab@linux-m68k.org>
> >> Cc: Eli Zaretskii <eliz@gnu.org>
> >> Date: Sun, 14 Mar 2021 08:28:36 +0100
> >> 
> >> >  #0  0x70f5ac3e in libgccjit-0!_Z17gimple_build_callP9tree_nodejz ()
> >> 
> >> Is this prefix part of the symbol name?  A valid mangled name must start
> >> with _Z.
> >
> > No, that's the name of the shared library.  This is a known
> > peculiarity on Windows, see "Non-debug DLL Symbols" in the GDB manual.
> 
> So it _is_ part of the symbol name?

AFAIU, it's how the symbol is recorded in minsyms, yes.